Multi-Environment Screening of Durum Wheat Genotypes for Drought Tolerance in Changing Climatic Events

: Durum wheat is the most widely grown cereal in Tunisia, but its production is threatened by drought, which is exacerbated by climate change. This study aimed to identify drought-tolerant durum wheat genotypes from ﬁve modern varieties and six landraces in a multi-environment trial at two sites (Kef and Siliana, Tunisia) during three growing seasons under rainfed and irrigated conditions. Six drought tolerance indices (mean productivity (MP), geometric mean productivity (GMP), stress susceptibility index (SSI), tolerance index (TOL), stress tolerance index (STI), and yield stability index (YSI)) were used to evaluate the 11 genotypes. The environment was the dominant source of variation for grain yield (GY; 94.27%), followed by the environment × genotype interaction (4.06%) and genotype (1.65%). Cluster analysis based on GY identiﬁed four environment-based groups with distinct water treatments, extreme minimum/maximum temperatures, and rainfall. Principal component analysis and a correlation matrix revealed that drought tolerance indices signiﬁcantly correlated with GY in non-stressed and stressed conditions and could be separated into four groups. Based on STI, MP, and GMP, G6 and G8 (landraces) were the most drought-tolerant genotypes attaining high GY in both conditions. TOL was able to discriminate G1, G3, and G5 (modern varieties) as well as drought-susceptible genotypes, all of which were suitable for irrigation. Genotypes G7, G9, G10, and G11 (landraces), which had high SSI and lowest STI, MP, GMP, and YSI values, were susceptible to drought and were thus not suitable for cultivation in both conditions. Finally, G2 and G4 (modern varieties), which had an intermediate rank for different indices, were classiﬁed as semi-tolerant or sensitive genotypes. Drought tolerance indices and genotype ranks were and G8), or genotypes with the ability to adapt (modern varieties G1, G3, and G5) to irrigated conditions.


Introduction
Among cereal crops, durum wheat (Triticum durum Desf.) is the 10th most commonly cultivated cereal worldwide and one of the most important food crops in the Mediterranean Rim with an important role in the human diet [1,2]. However, its production is threatened by climate change and extreme weather events [3][4][5]: North Africa, including Tunisia, is recognized as a climate change hotspot region [6][7][8] and is, therefore, particularly vulnerable to drought stress [9][10][11][12][13], which limits growth, development, and crop yield [14][15][16][17]. By 2030, Tunisia is expected to suffer from an annual average increase in temperature of 1.1 • C and thus an annual acute decrease in precipitation and water resources [6]. This will acutely influence rainfed durum wheat, which is the most important cultivated crop in Table 1. Codes, names, type, origin, and release date of 11 durum wheat genotypes used in this study.

Field Experiment and Crop Management
A field experiment was carried out in a split-plot design with three replicates per treatment (n = 3) during 2014-2015, 2015-2016, and 2016-2017 cropping seasons in two semi-arid regions of Tunisia, Kef and Siliana (Table S1). The treatment was defined by two water conditions: irrigated and non-irrigated surface. For each site, three blocs (190 m 2 ) per treatment (i.e., irrigated and non-irrigated) were subdivided into 11 plots of 2 m 2 (in total, 66 plots) and consisted of five rows that were 2 m long with a 0.2 m inter-row spacing and a 0.5 m inter-plot spacing. The distance between control and treated blocs was 10 m. Seeds were hand sown at a rate of 350 grains m −2 .

Water Treatments
For the non-irrigated treatment (i.e., control, rainfed, or stressed conditions), plants only received water from natural rainfall. In the well-watered treatment (i.e., irrigated or non-stressed conditions), plants received a specific amount of water for irrigation, in addition to rainwater. Irrigation water in the latter was supplied by a sprinkler. The amount of water that was provided was determined in accordance to the needs of the crop using CROPWAT 8.0 software, which considered rainfall distribution over each growing season at each site (Table S2). Information about the tested environments is provided in Table 2.

Indices Formula Unit References
Mean productivity (MP) MP = Ys+Yp 2 kg m −2 Rosielle and Hamblin [29] Geometric mean productivity (GMP) GMP = (Ys) × (Yp) kg m −2 Fernandez [28] Stress susceptibility index (SSI) SSI =  [31] Yp and Ys are the mean yield of genotypes under irrigated and rainfed conditions, respectively. Yp and Ys are the mean yield of all genotypes under irrigated and rainfed conditions, respectively.

Data Analysis
An analysis of variance (ANOVA) of the data for each trait was computed using R statistical software version 4.0 (The R Foundation for Statistical Computing). Significant differences between means were determined by Duncan's multiple range test (p ≤ 0.05, p ≤ 0.01, and p ≤ 0.001). Pearson's correlation coefficient and principal component analysis (PCA) were also calculated.

Environment Assessment
The amount (mm) and distribution of precipitation, as well as mean temperature ( • C) during the three durum wheat-growing seasons at both sites, were recorded from the national weather stations in Kef and Siliana ( Figure 1). Precipitation (mm) and extreme temperature events (low and high, • C) during the most sensitive period of durum wheat growth (i.e., the reproductive stage, which occurs in March, April, and May) from 2007 to 2017 at Kef and Siliana sites are presented in Figures 2 and 3.

Environment Assessment
The amount (mm) and distribution of precipitation, as well as mean temperature (°C) during the three durum wheat-growing seasons at both sites, were recorded from the national weather stations in Kef and Siliana ( Figure 1). Precipitation (mm) and extreme temperature events (low and high, °C) during the most sensitive period of durum wheat growth (i.e., the reproductive stage, which occurs in March, April, and May) from 2007 to 2017 at Kef and Siliana sites are presented in Figures 2 and 3.       At both sites, rainfall and temperature (spring heat and frost shock) in each cropping season differed, affecting the performance of genotypes under rainfed and irrigated conditions. On observing the 10-year (2007-2017) climatic data, the 2014-2015 cropping season was characterized by, on the one hand, a precipitation record (i.e., April had the lowest rainfall amount (0 mm) at both sites), and, on the other hand, extreme minimum tem-  At both sites, rainfall and temperature (spring heat and frost shock) in each cropping season differed, affecting the performance of genotypes under rainfed and irrigated conditions. On observing the 10-year (2007-2017) climatic data, the 2014-2015 cropping season was characterized by, on the one hand, a precipitation record (i.e., April had the lowest rainfall amount (0 mm) at both sites), and, on the other hand, extreme minimum temperatures (−1. ANOVA revealed significant (p < 0.01) effects of environments and genotypes on all agronomic traits. Except for SN and GY, the environment × genotype interactions were significant (p < 0.05, p < 0.01, and p < 0.001) for almost all traits (Table 4). In addition, combined ANOVA results showed that 94.27% of the variation in GY was explained by the effect of environments followed by the environment × genotype interaction (4.08%) and by the genotype variation (1.65%; Table S3).

Genotypic Grain Yield Performance and Drought Tolerance Selection
The GY of each durum wheat genotype under the different tested environments is shown in Table 5. In the 2014-2015 cropping season, when frost shock and early heatwaves were experienced, the highest mean GY was obtained by G6 in Kef and by G1 and G4 in Siliana in the irrigated treatment. In the 2015-2016 season, the three top-yielding genotypes in the irrigated treatment were the modern varieties G2, G5, and G1 at Kef, and G5, G1, and G3 in Siliana, but these did not differ significantly from other genotypes. Under stressed conditions, the highest GY was obtained for modern variety G5 at Siliana. Despite this, genotypes showed the same trends in Kef. In the last cropping season (2016-2017), modern varieties G1 and G3 had the highest GY, while landrace G11 displayed the lowest GY (0.34 kg m −2 ) under favorable conditions in Siliana. Under water stress, landrace G6 and modern variety G1 showed the greatest GY in Kef and Siliana, respectively. Over the 12 environments, the highest GY was observed for G1, G3, and G6 (0.29 kg m −2 ) and the lowest GY for G11 (0.23 kg m −2 ).

Genotypic Grain Yield Performance and Drought Tolerance Selection
The GY of each durum wheat genotype under the different tested environments is shown in Table 5. In the 2014-2015 cropping season, when frost shock and early heatwaves were experienced, the highest mean GY was obtained by G6 in Kef and by G1 and G4 in Siliana in the irrigated treatment. In the 2015-2016 season, the three top-yielding genotypes in the irrigated treatment were the modern varieties G2, G5, and G1 at Kef, and G5, G1, and G3 in Siliana, but these did not differ significantly from other genotypes. Under stressed conditions, the highest GY was obtained for modern variety G5 at Siliana. Despite this, genotypes showed the same trends in Kef. In the last cropping season (2016-2017), modern varieties G1 and G3 had the highest GY, while landrace G11 displayed the lowest GY (0.34 kg m −2 ) under favorable conditions in Siliana. Under water stress, landrace G6 and modern variety G1 showed the greatest GY in Kef and Siliana, respectively. Over the 12 environments, the highest GY was observed for G1, G3, and G6 (0.29 kg m −2 ) and the lowest GY for G11 (0.23 kg m −2 ).   Table 1); 3 E: environment, which is a combination of site, years, and water treatment (see Table 2); 4 means with similar letter(s) in each trait are not significantly different at p ≤ 0.05 (Duncan's multiple range test).
DTIs and genotype ranks are provided in Table 6. A significant effect of years on GY was observed, hence the variation in genotype ranks from year to year. Thus, durum wheat genotypes were screened for drought tolerance each year. The highest STI, MP, and GMP were observed for G6, followed by G3 and G8 in the first cropping season (2014-2015), and for G5 and G8 in the second season (2015-2016). In contrast, G1 and G3 performed best under stressed and non-stressed conditions in the third cropping season (2016-2017). Lowest TOL was observed for G11 and G8, G6 and G8, and G11 and G6 during the first, second, and third cropping seasons, respectively. According to this index, these genotypes experienced the least reduction in GY under rainfed conditions and were more resistant to drought stress. SSI displayed a similar ranking pattern as YSI. Lowest SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season.
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP seem to be the most effective parameters to select drought-tolerant genotypes in both conditions. TOL, which was significantly and positively correlated with Yp (r = 0.75; p < 0.01) and significantly and negatively correlated with Ys (r = −0.32; p < 0.01), might be useful to select genotypes with large GY, but only in the stressed environment.  SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 2 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 3 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 4 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 5 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 6 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 7 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 8 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season. ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 11 SSI and highest YSI were observed in G8 and G2 in the first cropping season, G8 and G6 in the second cropping season, and G6 and G7 in the third cropping season.   ; 1 STI: stress rance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress sustibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1).
To identify a suitable index for the drought tolerance selection, the correlation coefficient between different DTIs and GY in both conditions was determined (Table 7). Results indicate a positive and significant correlation between Yp and STI (r = 0.35; p < 0.01), MP (r = 0.89; p < 0.01), and GMP (r = 0.68; p < 0.01), and between Ys and STI (r = 0.88; p < 0.01), GMP (r = 0.93; p < 0.01), and MP (r = 0.76; p < 0.01). In addition, these indices had a positive and significant (p < 0.05) correlation between each other. Thus, STI, MP, and GMP ; 1 STI: stress tolerance index; 2 MP: mean productivity; 3 GMP: geometric mean productivity; 4 TOL: stress tolerance; 5 SSI: stress susceptibility index; 6 YSI: yield stability index; 7 G: genotypes (see Table 1). To better separate and classify the different genotypes, PCA was employed ( Figure 5). The first two PCAs explained 99.2% of the total variation in GY. The observed positive and significant correlation between STI, MP, and GMP, and between YSI and Ys indicated that these indices were able to discriminate G6 and G8 (Group I) as the stress-tolerant group with high GY in both conditions and good stability in the control conditions. Based on the positive and significant correlation between TOL and Yp, G1, G3, and G5 in Group II appeared to be susceptible to drought stress, although they performed well under irrigated conditions. Ys had a highly positive and significant correlation with STI, MP, GMP, and YSI and a significantly negative correlation with SSI. These last correlations indicated that G7, G9, G10, and G11 (Group III), which had the highest SSI, could be separated and were considered to be susceptible to drought and thus unsuitable in both conditions, especially unstable GY in the stressed conditions. G2 and G4 (Group IV) had an intermediate rank for the different indices, so they could be classified as semi-tolerant or sensitive genotypes. gated conditions. Ys had a highly positive and significant correlation with STI, MP, GMP, and YSI and a significantly negative correlation with SSI. These last correlations indicated that G7, G9, G10, and G11 (Group III), which had the highest SSI, could be separated and were considered to be susceptible to drought and thus unsuitable in both conditions, especially unstable GY in the stressed conditions. G2 and G4 (Group IV) had an intermediate rank for the different indices, so they could be classified as semi-tolerant or sensitive genotypes.

Discussion
Variation in wheat GY is strongly dependent on environmental conditions [38,39]. In this study, water treatment had a dominant effect on GY, in addition to the effect of the cropping season, which was significant due to climatic changes. Variation in annual weather conditions can affect the degree of stress experienced by a crop [40]. In our study, rainfall (amount and distribution) and temperature varied considerably from year to year, especially during critical periods of crop development (anthesis and grain filling), as was recorded in the first and the third cropping seasons, where a record lowest amount of rainfall and extreme minimum temperatures were registered (Figures 2 and 3). Therefore, four environmental groups were identified based on drought and temperature stresses ( Figure 4). Under well-watered conditions, GY was significantly correlated to GN/S (r = 0.63; p < 0.01), BY (r = 0.4; p < 0.01), and TKW (r = 0.20; p < 0.01; Table S4). Nonetheless, GY significantly correlated with all agro-morphological traits (i.e., PH, SL, GN/S, SN/m 2 , TKW, and BY) under water stress. In the same conditions, the lowest PH, SL, GN/S, SN, TKW, BY, and GY were observed for all durum wheat genotypes in the third environment group (EG3) due to a deficit in rainfall. Similar findings were obtained by Maqbool et al. [41] and Liu et al. [42], who reported that drought stress significantly reduced GY, PH, GN/S, and TKW in bread wheat. According to Dehgahi et al. [43], wheat GY positively correlated with annual rainfall, and its effect varies from year to year. In Tunisia, in semi-arid regions, the amount of precipitation was unable to cover water requirements for durum wheat, especially during March, April, and May when crop water needs are high, and the amount of rainfall played a crucial role in determining crop production [44]. That was the main reason why irrigation was necessary for Tunisian conditions. GY was also influenced by extreme temperature events. A review by Akter and Islam [45] revealed that heat stress reduces wheat grain number and size by reducing the grain-filling period and assimilate translocation. In addition, Shirdelmoghanloo et al. [46] reported that from 600 wheat field trials, 15% of the reduction in GY was due to temperatures exceeding 30 • C during or around flowering [47]. In our research, the lowest GY, its related traits (i.e., GN/S, SN, and TKW), and BY were obtained in the third growing season, which could be due to the combined effect of drought and heat stress. However, the joint impact of drought and heat stress on wheat growth and GY could be more pronounced than individually [48]. Together, these stressors reduced the number of plants/m 2 , PH, and BY/m 2 of rainfed wheat and barley [49] and GY in wheat by as much as 50% [50]. In general, this reduction could be explained by the greatest damage caused by frost and heat occurring during ear emergence and near anthesis. In all durum species, anthesis was the most sensitive stage when heat and frost shock caused sterility and the abortion of formed grains, and thus a significant loss of yield [51,52]. Heat and frost shock, which are serious limiting factors for GY, may increase during future climate change scenarios [51].
Based on the obtained results for GY, which was the main criterion for the selection of drought tolerance, important genetic variability was observed between genotypes. Genotypic variation in GY was greater (higher CV, %) in the rainfed environment than in the irrigated environment (lower CV, %; Table 5). Under stress conditions, large genotypic differences and responses were expected [53]. Thus, germplasm might be a useful genetic source in wheat breeding programs for developing drought tolerance. Similar results were obtained by Mwadzingeni et al. [21] in 88 bread wheat lines that were evaluated under greenhouse and field conditions during two growth seasons (four tested environments) in South Africa in order to determine the level of drought tolerance. In the present study, most modern varieties had the best response to supplemental irrigation. However, under rainfed conditions, landraces displayed the highest GY. Several studies showed that landraces were better adapted and tolerant to environmental stress, especially drought, compared to breeding lines, modern, and old varieties [40,[54][55][56]. Landraces were characterized by superior stem length, biomass production, and late flowering which cumulatively played a crucial role in the determination of GY. A longer stem, which resulted in a greater amount of transferred assimilates in the form of soluble carbohydrates [57], significantly contributed to grain filling [48]. Under water stress, in 18 durum wheat genotypes, the plant biomass had a positive effect on GY (r = 0.80) between tillering and maturity stages [58]. Late flowering and early heading by vigorous and rapid growth were important traits to avoid spring frost shock in durum wheat genotypes [51]. In addition, under a Mediterranean environment, where terminal drought frequently occurs, early heading was a defining characteristic to obtain high GY [59].
Based on the correlation matrix, cluster, and PCA analyses, three DTIs (STI, MP, and GMP) were highly correlated to each other and to GY in non-stressed and stressed conditions (Table 7), prompting us to use them to screen drought-tolerant genotypes with high GY in both conditions. Our findings were in agreement with findings by Mohammadi [40] in 24 durum wheat genotypes assessed over four years under rainfed and irrigated conditions. They found that the correlation between these indices was significant (p ≤ 0.01) at all stress levels (mild, moderate, and severe), which indicated that they were useful for ranking genotypes. In addition, Nouri et al. [60] and Subhani et al. [61] reported that these drought indices were preferred and useful to differentiate durum wheat and barley genotypes. In the present work, STI, MP, and GMP identified G6 and G8 (landraces) as the most drought-tolerant genotypes with high GY under irrigated and rainfed conditions. These genotypes have a good tolerance compared to G3, a drought-tolerant genotype. Thus, this group of genotypes had a good adaptation in a specific environment under drought conditions with irregular rainfall distribution, insufficient rainfall, as well as extreme temperature events. Otherwise, TOL was significantly and positively correlated with GY under optimal conditions but negatively correlated under rainfed conditions. TOL discriminated G1, G3, and G5 (modern varieties) as susceptible genotypes to drought but as suitable genotypes in irrigated conditions. The selection based on this index resulted in enhanced GY under favorable conditions [62]. Anwaar et al. [63] used SSI to identify drought-tolerant genotypes in 50 bread wheat landraces evaluated under rainfed and irrigated conditions and showed that tolerant genotypes could be selected by low SSI and TOL values. In this study, genotypes G7, G9, G10, and G11 (landraces), which had high SSI and lowest STI, MP, GMP, and YSI, were susceptible to drought and unsuitable genotypes for cultivation in both conditions.

Conclusions
The present study showed, on the one hand, a significant environmental effect on genotypic GY performance: the amount and distribution of rainfall, as well as extreme temperature events, were the main factors reducing durum wheat growth and GY in semi-arid environments. On the other hand, DTIs and genotype ranks served as helpful tools to screen drought-tolerant genotypes under a range of environments. Our results revealed that DTIs were significantly correlated with GY under non-stressed and stressed conditions. Landraces G6 and G8 were the most drought-tolerant genotypes, attaining high GY in both conditions. However, modern varieties (G1, G3, and G5), which were drought-susceptible genotypes, were only suitable for irrigation. Wide genotypic variation offered a tremendous opportunity for further crop improvement to drought as well as extreme minimum and extreme maximum temperature stresses.

Conflicts of Interest:
The authors declare no conflict of interest.